In recent years, there has been a technique of recognizing character strings by acquiring data of sound collected by a microphone (hereinafter, also simply referred to as “collected sound data”) and performing speech recognition on the collected sound data. For example, an example of a method of constructing context acoustic models for speech recognition has been disclosed (see Patent Literature 1, for example).